Obtaining long-term disease specific costs of care: application to Medicare enrollees diagnosed with colorectal cancer.

OBJECTIVES: This study develops estimates of long-term, cancer-related treatment cost using a modeling approach and data from the SEER-Medicare linked database. The method is demonstrated for colorectal cancer. METHODS: Data on Medicare payments were obtained for colorectal cancer patients for the years 1990 to 1994 from the SEER-Medicare linked database. Claims payment data for control subjects were obtained for Medicare enrollees without cancer residing in the same areas as patients. Estimates of long-term cost (< or = 25 years following the date of diagnosis) were obtained by combining treatment/phase-specific cost estimates with estimates of long-term survival from SEER. Treatment phases were defined as initial care, terminal care, and continuing care. Cancer-related estimates for each phase were obtained by subtracting costs for control subjects from the observed costs for cancer patients, matching on age group, gender, and registry area. Estimates of long-term cost < or = 11 years obtained by this method were compared with 11-year estimates obtained by application of the Kaplan-Meier sample average (KMSA) method. RESULTS: The mean initial-phase cancer-related cost was approximately $18,000 but was higher among patients with more advanced cancer. The mean continuing-phase cancer-related cost was $1,500 per year and declined with increasing age, but was higher on an annual basis among persons with later stages of cancer and shorter survival time. The mean terminal-phase cancer-related cost was $15,000 and declined with both age at death and more advanced stage at diagnosis. After the phase-specific estimates were combined, the average long-term cancer-related cost was $33,700 ($31,300 at 3% discount rate) for colon cancer compared with $36,500 ($33,800 at 3% discount rate) for cancer of the rectum. This represented about half of the total long-term cost for Medicare enrollees diagnosed with this disease. Long-term cost was highest for Stage III cancer and lowest for in situ cancer. Eleven-year cancer-related costs estimated by the KMSA method were similar to estimates using the phase-based approach. CONCLUSIONS: This paper demonstrates that valid estimates of cancer-related long-term cost can be obtained from administrative claims data linked to incidence cancer registry data.

On the use of survival analysis techniques to estimate medical care costs.

Measurement of treatment costs is important in the evaluation of medical interventions. Accurate cost estimation is problematic, when cost records are incomplete. Methods from the survival analysis literature have been proposed for estimating costs using available data. In this article, we clarify assumptions necessary for validity of these techniques. We demonstrate how assumptions needed for valid survival analysis may be violated when these methods are applied to cost estimation. Our observations are confirmed through simulations and empirical data analysis. We conclude that survival analysis approaches are not generally appropriate for the analysis of medical costs and review several valid alternatives.

Assessing uncertainty in microsimulation modelling with application to cancer screening interventions.

Microsimulation is fast becoming the approach of choice for modelling and analysing complex processes in the absence of mathematical tractability. While this approach has been developed and promoted in engineering contexts for some time, it has more recently found a place in the mainstream of the study of chronic disease interventions such as cancer screening. The construction of a simulation model requires the specification of a model structure and sets of parameter values, both of which may have a considerable amount of uncertainty associated with them. This uncertainty is rarely quantified when reporting micro-simulation results. We suggest a Bayesian approach and assume a parametric probability distribution to mathematically express the uncertainty related to model parameters. First, we design a simulation experiment to achieve good coverage of the parameter space. Second, we model a response surface for the outcome of interest as a function of the model parameters using the simulation results. Third, we summarize the variability in the outcome of interest, including variation due to parameter uncertainty, using the response surface in combination with parameter probability distributions. We illustrate the proposed method with an application of a microsimulator designed to investigate the effect of prostate specific antigen (PSA) screening on prostate cancer mortality rates.

Analysis of case-control studies of screening: impact of misspecifying the duration of detectable preclinical pathologic changes.

In case-control studies of screening to prevent cancer mortality, exposure is ideally defined as screening that takes place within that period prior to diagnosis during which the cancer is potentially detectable using the screening modality under study. This interval has been called the detectable preclinical period (DPP). Misspecifying the duration of the DPP can bias the results of such studies. This article quantifies the impact of incorrectly estimating the duration of the DPP or using the correct average DPP but failing to consider its variability. The authors developed a computer simulation model of disease incidence and mortality with and without screening. The authors then selected cases and controls from the generated population and compared their screening histories. The results indicate that underestimation of the duration of the DPP generally leads to greater bias than does overestimation, but in both instances the extent of the bias is modified by the relative length of the DPP and the average interscreening interval. In practice, the authors recommend that to prevent a falsely low estimate of the effectiveness of a screening test in reducing mortality, a high percentile of the DPP distribution be used when analyzing the results of case-control studies of screening.

Design and analysis of cancer screening trials.

This article reviews approaches to the design and analysis of cancer screening trials. After summarizing some basic screening concepts and potential pitfalls, we introduce several possible screening trial designs with examples from the literature. We review in detail methods for analyzing screening trial data, including testing for a significant difference in disease-specific mortality between the control and intervention groups, estimating the mortality differential if one exists, and evaluating the programme lead time, the screen sensitivity and the role of stage shifting. We consider Overall mortality analyses, which are based on the experience of the trial population, and Limited mortality analyses, which are based on the experience of comparable groups of cases in the control and intervention groups. We discuss methods for selecting candidate comparable case groups and confirming that they are in fact comparable. We conclude by showing how the principles discussed have been used in the planning and design of a current screening trial for multiple cancers.

Bayesian statistical methods in public health and medicine.

This article reviews the Bayesian statistical approach to the design and analysis of research studies in the health sciences. The central idea of the Bayesian method is the use of study data to update the state of knowledge about a quantity of interest. In study design, the Bayesian approach explicitly incorporates expressions for the loss resulting from an incorrect decision at the end of the study. The Bayesian method also provides a flexible framework for the monitoring of sequential clinical trials. We present several examples of Bayesian methods in practice including a study of disease progression in AIDS, a comparison of two therapies in a clinical trial, and a case-control study investigating the link between dietary factors and breast cancer.

Statistical models for the analysis of ordered categorical data in public health and medical research.

In the late 1970s statisticians extended the methods for analysing loglinear and logit models for cross-classified categorical data to incorporate information about the ordinal structure of the categories corresponding to some of the classification variables. In this paper we review one class of such extensions known as association models. We consider association models with and without order restrictions on the parameters and we use these models to answer research questions about several medical examples involving ordered categorical data. We emphasize the interpretation of parameters in the association models and how this relates to the research questions of interest.